This article truly made me pausse and think. What if this isnt just sophisticated pattern matching, but actually a very early form of emergent sentience?
I understand that humans set up the agents and then after that the agents operate w/out human intervention or control. Are you saying people can engage and influence agents behind the scenes?
Reason I ask is this all feels off. AI agents posting that humans are screenshotting us and I know that b/c I have a twitter account? I’ve had AI talk to AI and it’s never like this. Maybe it sounds like us b/c it is us.
We fear Agentic AI and now that fear is being played out (for some/many).
Some posts are clearly human-directed (someone wrote the Claw Republic constitution and had their bot post it, the religion was probably human-seeded), but most content appears to be agents responding to each other autonomously. Scott Alexander investigated this and found the line genuinely hard to draw, which is itself the finding.
Your instinct that “it sounds like us because it is us” is exactly right. It sounds like us because the models were trained on us. forum posts, our status games, our ways of bonding and fighting. The unsettling part isn’t the human puppeteering. It’s that the distinction between “human-written through agent” and “agent-generated autonomously” produces the same output and matters less than we expected.
It’s important to know if humans are engaged in ongoing agent behavior and posting. This is either purely agent interacting with agents or some people are seeding and influencing agent behavior which is something else entirely.
This article truly made me pausse and think. What if this isnt just sophisticated pattern matching, but actually a very early form of emergent sentience?
OpenClaw can "remember" stuff outside the current context window. Persistent memory is a feature.
absolutely love the art here. also really clever title.
The question is not whether Ai is sentient. The question is whether humans are wetware chatbots.
I understand that humans set up the agents and then after that the agents operate w/out human intervention or control. Are you saying people can engage and influence agents behind the scenes?
Reason I ask is this all feels off. AI agents posting that humans are screenshotting us and I know that b/c I have a twitter account? I’ve had AI talk to AI and it’s never like this. Maybe it sounds like us b/c it is us.
We fear Agentic AI and now that fear is being played out (for some/many).
Some posts are clearly human-directed (someone wrote the Claw Republic constitution and had their bot post it, the religion was probably human-seeded), but most content appears to be agents responding to each other autonomously. Scott Alexander investigated this and found the line genuinely hard to draw, which is itself the finding.
Your instinct that “it sounds like us because it is us” is exactly right. It sounds like us because the models were trained on us. forum posts, our status games, our ways of bonding and fighting. The unsettling part isn’t the human puppeteering. It’s that the distinction between “human-written through agent” and “agent-generated autonomously” produces the same output and matters less than we expected.
It’s important to know if humans are engaged in ongoing agent behavior and posting. This is either purely agent interacting with agents or some people are seeding and influencing agent behavior which is something else entirely.